Super-Reduced Polyoxometalates: Excellent Molecular Cluster

Jun 2, 2014 - On the basis of structural, electron density, and molecular orbital studies, we present evidence that the super-reduction is accompanied...
3 downloads 10 Views 4MB Size
Article pubs.acs.org/JACS

Super-Reduced Polyoxometalates: Excellent Molecular Cluster Battery Components and Semipermeable Molecular Capacitors Yoshio Nishimoto,† Daisuke Yokogawa,†,‡ Hirofumi Yoshikawa,†,§ Kunio Awaga,†,§ and Stephan Irle*,†,‡ †

Department of Chemistry, Graduate School of Science, Nagoya University, Nagoya 464-8602, Japan Institute of Transformative Bio-Molecules (WPI-ITbM), Nagoya University, Nagoya 464-8602, Japan § Research Center for Materials Science, Nagoya University, Nagoya 464-8602, Japan ‡

S Supporting Information *

ABSTRACT: Theoretical investigations are presented on the molecular and electronic structure changes that occur as αKeggin-type polyoxometalate (POM3−) clusters [PM12O40]3− (M = Mo, W) are converted toward their super-reduced POM27− state during the discharging process in lithium-based molecular cluster batteries. Density functional theory was employed in geometry optimization, and first-principles molecular dynamics simulations were used to explore local minima on the potential energy surface of neutral POM clusters adorned with randomly placed Li atoms as electron donors around the cluster surface. On the basis of structural, electron density, and molecular orbital studies, we present evidence that the super-reduction is accompanied by metal−metal bond formation, beginning from the 12th to 14th excess electron transferred to the cluster. Afterward, the number of metal−metal bonds increases nearly linearly with the number of additionally transferred excess electrons. In α-Keggin-type POMs, metal triangles are a prominently emerging structural feature. The origin of the metal triangle formation during super-reduction stems from the formation of characteristic three-center two-electron bonds in triangular metal atom sites, created under preservation of the POM skeleton via “squeezing out” of oxygen atoms bridging two metal atoms when the underlying metal atoms form covalent bonds. The driving force for this unusual geometrical and electronic structure change is a local Jahn−Teller distortion at individual transition-metal octahedral sites, where the triply degenerate t2 d orbitals become partially filled during reduction and gain energy by distortion of the octahedron in such a way that metal−metal bonds are formed. The bonding orbitals show strong contributions from mixing with metal−oxygen antibonding orbitals, thereby “shuffling away” excess electrons from the cluster center to the outside of the cage. The high density of negatively charged yet largely separated oxygen atoms on the surface of the super-reduced POM27− polyanion allows the huge Coulombic repulsion due to the presence of the excess electrons to be counterbalanced by the presence of Li countercations, which partially penetrate into the outer oxygen shell. This “semiporous molecular capacitor” structure is likely the reason for the effective electron uptake in POMs.



INTRODUCTION

either incorporated into POM structures themselves or encapsulated within them.8 In recent years, the number of experimental9−12 as well as theoretical13−18 investigations of various POM clusters has increased because of their fascinating molecular and electronic structures and properties. However, the presence of a minimum of six transition-metal atoms in even the smallest possible POM anions represents a challenge for substantive quantum-chemical studies.8 Fortunately, in recent years, reasonably accurate density functional theory (DFT) has become applicable to such large and complicated molecular systems. Thus, thanks to combined experimental and theoretical efforts, a variety of POM properties such as catalytic activity,9,10,16 single-molecule magnetism,11 and electrochemical redox potentials12,18−20 are now fairly well understood.

Polyoxometalate (POM) was first reported by Jöns Jacob Berzelius in 1826,1,2 and it was later found that this substance contains the [PMo12O40]3− anion. However, its precise molecular structure remained unknown for more than 100 years, until James F. Keggin resolved the structure of a related heteropolyacid species, H3[PW12O40]·nH2O, using X-ray spectroscopy in 1933.3 His data indicated that a central PO43− tetrahedral group is surrounded by 12 fused WO6 octahedrons sharing oxygens at their edges or vertices and that the trianion adopts Td symmetry. It later was confirmed that virtually all [XM12O40]x−-type POMs (X = P, Si, Al, etc.; M = Mo, W, V, etc.) feature this so-called α-Keggin structure. Other relevant POM structures, for instance, the smaller homopoly Lindqvist structure4 ([M6O19]2−) and larger heteropoly structures such as the Anderson−Evans 5,6 ([XM 6 O 24 ] x− ) and Dawson 7 ([X2M18O62]x−) structures, have been reported as well. With only a few exceptions, almost all chemical elements have been © XXXX American Chemical Society

Received: April 1, 2014

A

dx.doi.org/10.1021/ja5032369 | J. Am. Chem. Soc. XXXX, XXX, XXX−XXX

Journal of the American Chemical Society

Article

One of the most intriguing properties of POM clusters is their highly unusual capability to accept a large number of electrons,21−24 a property that finds its expression in the terms “electron reservoir”2 or “electron sponge”25 sometimes used in the literature. This means for instance that POM clusters have the potential to play an important role as a cathode-active material. In the mid-1970s, Jean Pierre Launay24 reported the preparation and characterization of electrochemically highly reduced metatungstate anions.2 On the basis of the observation of irreversible electrode reactions and simple molecular orbital (MO) considerations, he and his co-workers proposed that three joint octahedra constituting each of the four corners of the POM cluster would be reduced by up to six electrons and that this locally reduced structure would feature WIV−WIV bonds arranged in the form of triangles.24 However, these suggestions were mere speculation and lacked any direct evidence. More recently, our team reported that in high-capacity lithium batteries23,25−27 a Keggin POM anion containing 12 Mo atoms may indeed hold 24 excess electrons, as determined by in situobserved changes in the Mo ion average valences derived from Mo K-edge absorption edge energies of X-ray absorption nearedge structure (XANES) data.25 This high uptake of excess electrons (termed “super-reduction” in ref 25) is the current confirmed record for a single molecular cluster and is certainly remarkable since highly charged anions are usually prone to break covalent bonds and decompose spontaneously. Other chemical compounds exhibiting the features of super-reduction are Mnx26,27 and iron−sulfur28,29 clusters, although these compounds can accommodate only up to eight and four excess electrons per cluster, respectively. Thus, to achieve superreduction, it is necessary to use molecular clusters with a large number of transition-metal atoms. Since POMs are famous for their capability to form supramolecular condensates with welldefined structures,2 they possess tremendous potential for the design of molecular clusters capable of super-reduction, possibly even exceeding the current record of 24-electron super-reduction. In the present work, we focus on the question of how the Keggin-type POM cluster can reach a super-reduced state in a molecular cluster battery (MCB) without being ripped apart by the repulsive Coulombic forces due to the large number of excess electrons. In the extended X-ray absorption fine structure (EXAFS) spectrum reported by Wang et al.,25 three unique structural changes were reported during reduction by up to 24 electrons: (i) most notably, the overall Mo−Mo distances shrank from about 3.6 Å in the “native” Keggin [PMo12O40]3− cluster trianion30 to about 2.6 Å; (ii) the Mo−O bonds pointing outward away from the cluster [the Mo−O (out) bonds; see Figure 1] increased in length from 1.7 to about 1.9 Å; and (iii) the inner Mo−O [Mo−O (in)] bonds were shortened from 2.4 to about 2.0 Å. To better understand these structural changes recorded in situ by EXAFS, we performed our study theoretically by means of DFT calculations using both static and first-principles molecular dynamics (FPMD) approaches. For the discussion of changes in molecular and electronic state structure during reduction, we combined available experimental and theoretical data and employed MO analyses in combination with ligand field theory.31 Our approach models the super-reduced state of POM by supplying explicit alkali-metal atoms around the POM cluster in the DFT calculations.

Figure 1. BP86-D-optimized structure of [PMo12O40]3− and definitions of characteristic bond lengths. The green-, red-, and olive-colored spheres represent molybdenum, oxygen, and phosphorus atoms, respectively.



COMPUTATIONAL DETAILS

The present study mainly employed the heteropolymolybdate αKeggin polyoxometalate cluster [XMo12O40]3− (X = P) as the model system, and we refer to this particular Keggin cage as “POM” in the remainder of the paper. Several calculations were also carried out for its tungsten analogue [XW12O40]3− (X = P), and in these cases we refer to the Keggin cage as “W-POM”. All of the molecular and electronic structure calculations as well as the on-the-fly direct FPMD simulations were carried out using DFT32,33 as implemented in the TURBOMOLE code.34 We selected the BP86 exchange−correlation35,36 and B3LYP37 hybrid functionals with the all-electron, polarized split-valence-type def-SV(P) basis set38 [hereafter abbreviated as SV(P)] for Li, O, Na, and P atoms and the TURBOMOLE standard effective core potentials (ECPs) for Mo and W.39 The semiempirical dispersion correction as formulated by Grimme40 was introduced in order to describe long-range van der Waals interactions, and the functionals with this correction are denoted with a “-D” suffix. A geometry optimization using a larger valence triple-ζ-quality basis set (TZVPP)41 for all elements was also performed for the “native” POM3− cluster. For all of the BP86 DFT calculations, we introduced the resolution of identity (RI) approximation42 in combination with optimized auxiliary basis sets,38,39,41,43 which significantly enhanced the speed of calculation with negligible loss of accuracy. No symmetry constraints were applied in the calculations if not explicitly mentioned otherwise. In all of the calculations we assumed a closed-shell singlet state and used spin-restricted MOs. For a few selected cases, we computed energies of electronic states with higher spin multiplicities and found that they were usually higher than singlet energies, indicating that the system tries to adopt a low spin electronic ground state whenever possible. The calculations on the POM3− cluster were carried out in the absence of explicit counterions (i.e., assuming three negative charges on this cluster) unless indicated otherwise. In order to model the reduced POM(3+n)− clusters, we added 3 + n neutral alkali-metal atoms (either lithium or sodium) around a neutral POM cluster and set the total charge of the system in the DFT calculation to zero. For the FPMD simulations in the canonical NVT ensemble, the environmental target temperature was controlled by the Nose−Hoover chain thermostat44−46 as implemented in the TURBOMOLE code (program “frog”) The target temperature was set to 500 K. The thermostat relaxation time was 12.094 fs and the time integration interval was 1.935 fs, since no light hydrogen atoms are included in the system. For MO-based population analyses, we employed the AOMix47,48 code, and analysis of the charge densities was carried out on the basis of natural population analysis (NPA)49 as well as Wiberg bond orders.50 For visualization of molecular and electronic structures, the VMD51 and gOpenMol52,53 software packages were used. B

dx.doi.org/10.1021/ja5032369 | J. Am. Chem. Soc. XXXX, XXX, XXX−XXX

Journal of the American Chemical Society



Article

RESULTS AND DISCUSSION Molecular Structure of the Native Keggin POM3−. The initial geometry of the heteropolymolybdate POM3− (X = P) cluster was taken from the X-ray structure of Liu et al.30 We subjected this geometry to structural optimization at the BP86D/SV(P) and B3LYP-D/SV(P) levels of theory in vacuum without explicit counterions, assuming a total charge of −3. Table 1 displays key structural parameters for the crystal

confirming a general trend in the performance of various DFT functionals for transition-metal-containing systems.55 We additionally performed a BP86-D calculation with a larger basis set (TZVPP) to validate our general use of the double-ζquality SV(P) basis set. The geometry was similar to the SV(P)-optimized result, and the corresponding RMSD value of this structure with respect to the X-ray geometry was 0.134 Å, indicating that our results did not have significant basis set dependence in the prediction of geometries. Overall, the trend of DFT gas-phase optimizations toward larger cage structures is indicative of the important role of crystal field effects on POM3− molecular cluster geometries. On the basis of these results, we therefore decided to employ the BP86-D method for the study of POM redox processes because of its reasonable performance and low computational cost. Strategies for Modeling the Molecular and Electronic Structures of Reduced POMs. The quantum-chemical study of anions is notoriously difficult, and to make matters worse, even though multiply charged anions, such as SO42−, CO32−, and PO43−, are ubiquitous in chemistry, they are not stable with respect to autodetachment as isolated species in the gas phase.56 Clearly, to accurately describe polyanionic species theoretically, it would be required to (a) accurately account for electron correlation, (b) use sufficiently large basis sets to accommodate the diffuse electron density of negatively charged species, and (c) include the environment of the molecular polyanion in the quantum-chemical calculation. Presently none of the available quantum-chemical methodologies are capable of “precisely” taking into account any of these three important theoretical requirements, let alone all three of them together. Thus, any attempt to model a super-reduced POM molecular cluster has to be conducted within the current limits of computational feasibility, and solid experimental observations must be used to compensate for methodological deficiencies. Since the Keggin POM contains 12 transition-metal atoms plus 40 oxygen atoms, ab initio methods such as second-order Møller−Plesset perturbation theory57 or configuration interaction58 are out of the question for the treatment of electron correlation effects. DFT is presently the only computationally feasible method here, and we have to accept its predictions without a thorough check by more accurate methods and with only limited complementary information from EXAFS spectra. In terms of the basis set, as we have already indicated above, the split-valence polarized basis set plus ECPs for the transition metals is sufficiently accurate to describe the geometry of the POM3− cluster. Since our approach of modeling the superreduced state involves the calculation of charge-neutral species by way of inclusion of explicit counterions, the use of larger basis sets is not expected to qualitatively change the BP86-D/ SV(P) results of our investigations, in particular since the large number of diffuse basis functions from the alkali-metal atoms provides additional flexibility in the description of the MOs of the super-reduced POMs. Finally, the atomistic structure of the POM cluster environment in the MCB during charging and discharging is presently unknown, and we thus relied on the assumption that close contacts between the reduced POMs and the countercations (Li+) and the dielectric field of the graphitic solid, surrounded by the ethylene carbonate solvent, stabilize the super-reduced species.25 The influence of the countercations is dominant among the interactions with the cluster environment, and for this reason we explicitly included them in all calculations on super-reduced POMs. However, estimates of

Table 1. Comparison of Key Structural Parameters in the Xray30 and DFT-Optimized Structures of POM3− a Mo−Mo Mo−O (out) Mo−O (in) Mo−O (frameA1) Mo−O (frameA2) Mo−O (frameB1) Mo−O (frameB2) RMSDb

X-ray

BP86-D

B3LYP-D

3.563 1.684 2.442 1.920

3.606 1.719 2.452 1.882 2.011 1.867 1.995 0.127

3.609 1.703 2.461 1.855 2.031 1.843 2.016 0.152

1.915 0.000

a

All values are given in units of Å. The locations of the bond types are given in Figure 1. bRMSD values of BP86-D and B3LYP-D are calculated with respect to the X-ray structure as reference.

structure and the BP86-D- and B3LYP-D-optimized geometries along with root-mean-square deviations (RMSDs) of the theoretical geometries from the crystal structure. Although experimentally the metal−oxygen bond lengths for α-Keggin POMs are typically divided into three categories, we identified different types, consistent with a previous theoretical study by the Poblet group,17 namely Mo−O (out), Mo−O (in), Mo−O (frameA1), Mo−O (frameA2), Mo−O (frameB1), and Mo−O (frameB2). In Td symmetry, Mo−O (frameA1) is identical to Mo−O (frameA2) and Mo−O (frameB1) is identical to Mo− O (frameB2), but these pairs are in fact nonequivalent in lowersymmetry calculations, as will be discussed below. The Cartesian coordinates after the geometry optimizations are given in the Supporting Information (SI) and are labeled as structure 1 (BP86-D) and structure 2 (B3LYP-D) for POM and structure 3 (BP86-D) for W-POM. The RMSD of the Cartesian coordinates was defined as usual without mass weighting and measures structural deviations between maximally overlapping structures. It can be seen that the DFT bond lengths do not deviate from the X-ray structure by more than 0.1 Å. The largest deviation can be seen for the Mo−O (frame) bond lengths. The existence of different Mo−O (frame) bond lengths in POMs was reported previously in the theoretical study by Yan et al.17 They observed alternating bond length (ABL) distortions and found that the origin of these distortions, which lower the point-group symmetry from Td to chiral T, is a pseudo-Jahn−Teller vibronic instability. In Table 1, the ABL distortions are clearly visible in our computed geometries, obtained by geometry optimizations without symmetry restrictions. The averages over these four Mo−O types are 1.939 and 1.936 Å using BP86-D and B3LYP-D, respectively, in good agreement with the X-ray structure. In fact, the BP86-Dand B3LYP-D-optimized geometries are remarkably similar, even though B3LYP itself is typically more suitable for systems containing main-group elements.54 Indeed, the RMSD values indicate that the BP86-D structure is in slightly better agreement with the X-ray data than the B3LYP-D structure, C

dx.doi.org/10.1021/ja5032369 | J. Am. Chem. Soc. XXXX, XXX, XXX−XXX

Journal of the American Chemical Society

Article

structure 4 in the SI). We noticed a substantial rearrangement of the positions of the Li atoms during the relaxation of the structure. The final POM structure contained four significantly shortened Mo−Mo distances (less than 3.0 Å), three of which formed a triangular Mo site consisting of Mo−Mo bonds in the range from 2.65 to 2.77 Å with an average of 2.73 Å (see Figure 2A). These bond lengths are only slightly longer than the previously reported experimental bond length of 2.6 Å for the super-reduced POM from EXAFS,25 and they confirm that super-reduction induces Mo−Mo bond formation. In addition, we found that Mo−O (out) bonds for Mo atoms involved in Mo−Mo bonding had significantly stretched by ∼0.2 Å to 1.97 Å on average, while the Mo−O (in) bonds were compressed by a similar amount to 2.18−2.20 Å (see Figure 2B). These changes in Mo−O bond distances are also in agreement with experimental findings. The ABL related to Mo−O (frame1) and Mo−O (frame2) bonds vanished, and the Mo−O (frame) bonds in this geometry exhibited values between 2.13 and 2.23 Å. This makes the Mo−O (frame) bonds difficult to distinguish from significantly shortened Mo−O (in) bonds in the superreduced species. NPA49 carried out on this ad hoc optimized POM + 27 Li structure 4 indicated that the POM cluster had taken up only 21.2 electrons from the Li atoms, as estimated from the NPA point charge distribution. We performed similar calculations with modified initial geometries of Li atoms and obtained varying structures, sometimes with and sometimes without Mo triangles. The corresponding total energies of these POM + 27 Li isomers varied on a hundred kcal/mol energy scale as a result of the strong Coulombic interactions between the countercations and the negatively charged POM cluster. These explorative studies demonstrated that the potential energy surface (PES) of the POM + 27 Li system is very rough, featuring high barriers separating minimum-energy structures associated with Li migration and POM structural changes.

their number and positions are not available from experiment, and it can be expected that in MCBs large ion density fluctuations can easily occur. Our approach for adding Li+ or Na+ counterions in the DFT calculations therefore followed a somewhat heuristic approach. We note in passing that Fleming et al.59 theoretically investigated the effect of absorption of a cluster anion onto a gold surface using polarization-induced mirror charges, but the explicit use of alkali-metal counterions is more accurate for modeling of a Li-ion battery component than a mirror charge model. Since no detailed structural information for solid and liquid environmental compounds is available from the experiment, we chose to neglect their influence entirely. Geometry Optimizations of Super-Reduced POMs. At first we attempted to perform calculations with 27 Li atoms surrounding a neutral POM cluster starting from the X-ray structure of the POM3− Keggin trianion.30 The Li atoms were initially positioned randomly around the cage, sufficiently close for interaction. After a full geometry optimization of this system, we obtained the structure shown in Figure 2 (labeled as

Figure 2. Optimized structure of POM with 27 Li, labeled as structure 4 in the SI: (A) side view showing Mo−Mo distances; (B) top view of the triangular Mo site showing Mo−O (in), Mo−O (out), and Mo−O (frame) distances. Alkali-metal atoms have been omitted for clarity. See Figure 1 for the definition of colors.

Figure 3. Histograms of the bond-length fluctuations during the MD simulations of the POM cluster with (A) 27 and (B) 35 explicit Li atoms. As illustrated in the structure at the right, the bond types are color-coded as follows: Mo−Mo, pink; Mo−O (out), red; Mo−O (in), green; Mo−O (frame), blue. The vertical dotted line in (B) indicates the snapshot geometry at 1.113 ps that we selected for subsequent geometry optimization of the POM + 35 Li system. D

dx.doi.org/10.1021/ja5032369 | J. Am. Chem. Soc. XXXX, XXX, XXX−XXX

Journal of the American Chemical Society

Article

Figure 4. Selected snapshots from trajectory B at (A) 0.000 ps (the initial structure), (B) 1.113 ps, (C) 1.258 ps, and (D) 1.935 ps (at the end of the MD simulation). See Figure 1 for the definition of P, O, and Mo colors; white spheres represent Li atoms.

structure 6, and the origin of the ineffectiveness of the larger sodium atoms60 in reducing the POM cluster are discussed in the SI. We note that such a counterintuitive result also has been reported for instance in the difference of the solvent stabilization potentials of small, polar water molecules and larger, purely ionic liquid cations.61 To summarize, the simulations we had performed thus far hinted at a qualitative relationship between the number of excess electrons on the POM and the appearance of short Mo− Mo bonds. Since Li was apparently more effective in achieving POM reduction, we decided to add up to eight more Li atoms to the POM + 27 Li system. Geometry optimizations starting from the initial structure 5 after the addition of eight more Li atoms at random initial positions around the cluster resulted in a structure that featured a total of six short Mo−Mo bonds, one complete and one partial triangular Mo site, and one Mo−Mo single bond with a Wiberg bond order50 of 0.936 (structure 7 in the SI). We then proceeded with FPMD simulations of the POM + 35 Li system following the same “recipe” as described above (i.e., starting from the POM3− X-ray geometry and randomly supplying 35 Li atoms around this cluster). As before, we employed a target temperature of 500 K and performed the simulation for 1.935 ps, generating trajectory B. Figure 4 displays selected snapshot geometries representing the structural changes occurring in this trajectory. In this case, we observed that during the FPMD simulation the POM + 35 Li cluster gradually disintegrated after around 1.26 ps, as can be seen in the rotated view of the final geometry in Figure 4D, which shows completely dissociated oxygen bound to nearby Li. The tendency toward disintegration can also be seen in the time evolution of the Mo−O (frame) bond lengths in Figure 3B, which start to exceed distances of around 3.0 Å starting from 1.2 ps. Figure 5 displays the RMSD of the POM cluster atomic coordinates in trajectory B, except for Li atoms, with respect to the initial geometry (the POM3− X-ray geometry). At first, the RMSD value steeply increases as a sign of changes in the POM molecular structure and then relatively quickly stabilizes at around 0.5 Å up to 1.3 ps, just after the snapshot in Figure 4C. Afterward, gradual disintegration of the cluster is shown by a continuous increase in the RMSD. We conclude that it is difficult to accommodate 35 Li atoms around the Keggin POM cluster, at least at the high temperature of our MD simulation. For further geometry analysis, in this case we extracted a snapshot geometry before cage disintegration, namely, the structure at 1.113 ps (see Figure 4B, corresponding to the structure at the vertical dotted line in Figure 3B), and subjected it to geometry optimization. The resulting POM + 35 Li molecular structure (structure 8 in the SI) featured two triangular Mo sites, two single Mo−Mo bonds, and even one MoMo double bond. The NPA indicates a charge transfer of

Clearly, ad hoc geometry optimizations became trapped in local minima that were structurally related to the initial choice of Li atom placement. In order to investigate the PES more broadly and avoid entrapment close to the initial geometries, we decided to perform on-the-fly FPMD simulations with subsequent structure optimization (“quenching”) on the basis of BP86-D/SV(P) energies and gradients. Molecular Dynamics Simulations of Super-Reduced POMs. We adopted a relatively high temperature of 500 K for the FPMD simulations in order to more fully explore the associated PES. We note that although the MCB experiments had been carried out at room temperature,25 the local temperature close to the POM clusters might actually be closer to the target temperature of our MD simulation. At first, we subjected the previously described POM + 27 Li system to FPMD simulations starting from the POM3− X-ray geometry (trajectory A). Figure 3A displays structural changes during the entire length of the MD simulation traced by the bond lengths for Mo−Mo in pink, Mo−O (out) in red, Mo−O (in) in green, and Mo−O (frame) in blue. The FPMD simulation was performed for 1000 time integration steps (i.e., for a total of 1.935 ps). The bond lengths undergo dramatic changes, in particular during the first 0.5 ps, indicating that major structural changes in the cage induced by electron transfer from Li atoms to the POM cluster take place during this time period. Most notably, three to four short Mo−Mo distances around 2.5−2.8 Å quickly appear in this FPMD simulation, starting at around 0.39 ps. When a bond is formed between two Mo centers, it follows that their distances to other Mo centers must increase, and therefore, a large variation in Mo−Mo distances can be observed. The Mo−O (out) bonds significantly stretch immediately from 1.7 Å (characteristic of the optimized POM3− structure) to an average of 1.85 Å in the FPMD simulations. The Mo−O (in) and Mo−O (frame) bond lengths fluctuate with larger amplitudes around average values of 2.47 and 2.06 Å, respectively, indicating that the structural integrity of the POM cluster is compromised. The averages and amplitudes of these particular bond-length fluctuations appear to be converged after around 1 ps of simulation time. We then proceeded to perform a geometry optimization of the final trajectory snapshot at 1.935 ps, which resulted in a structure that features three short Mo−Mo bonds (